Posts

You can find here some material about the “Lexical Resources” course that I’m teaching in 2019-2020 to Master M2 TAL in Nancy.

There are also a few material related to the course about Deep Learning that I have taught at Ecole des Mines de Nancy (3rd year students), but only a small part of it.

Here is a selection of interesting ICLR-2020 papers to read:

• “Privacy-preserving Representation Learning by Disentanglement”
• “Toward Controllable Text Content Manipulation”
• “PROGRESSIVE LEARNING AND DISENTANGLEMENT OF HIERARCHICAL REPRESENTATIONS”
• “Analyzing Privacy Loss in Updates of Natural Language Models”
• “Learning from Positive and Unlabeled Data with Adversarial Training”
• “A Gradient-Based Approach to Neural Networks Structure Learning”
• “Federated Learning with Matched Averaging”
• “Efficient and Robust Asynchronous Federated Learning with Stragglers”
• “Classification-Based Anomaly Detection for General Data”
• “Distributed Training Across the World”
• “Adversarially learned anomaly detection for time series data”
• “The Secret Revealer: Generative Model Inversion Attacks Against Deep Neural Networks”
• “Differentially Private Meta-Learning”

I’ve been very happy to co-organize, with Claire Gardent, the Python4NLP summer school in Nancy, France, during the last week of August.

The school has been a wonderful moment, and I’d like to warmly thank all participants ! I hope we’ll get in touch again soon.

Our OLKi project is progressing, and our federated platform is little by little getting into shape: the first mock-up is online ! Once completed, it will be the first ActivityPub-powered scientific social network…

Stay tuned for a feature-rich platform by the end of 2019 !

mutt is a text-based mailer: when you have big mailboxes, it’s much much faster and efficient than standard web mail interfaces and heavy clients. But of course, because it’s text-based, it’s also less “nice, fancy and beautiful”. But I care much more about speed than design, so, here is my config file to use mutt with INRIA zimbra: set copy="yes" # Save sent mail on Zimbra set from="christophe.cerisara@inria.fr" # Your email set folder="imaps://zimbra. [Read More]

The scientific community is used to publish papers in the PDF format, which is nice to read on big screens, but is an abomination on smaller screens, especially smartphones and e-ink screens (epub readers, kindle…). All scientists have to read a lot of papers, actually so many that it’d be a relief to be able to read some while on the go, or on ebook readers under the sun. But converting from PDF to epub format is never satisfactory, when we only have access to the PDF and not the source of the paper. [Read More]